| 1 |
350 |
julius |
/////////////////////////////////////////////////////////////////////
|
| 2 |
|
|
//// ////
|
| 3 |
|
|
//// or1200_fpu_intfloat_conv ////
|
| 4 |
|
|
//// Only conversion between 32-bit integer and single ////
|
| 5 |
|
|
//// precision floating point format ////
|
| 6 |
|
|
//// ////
|
| 7 |
|
|
//// Author: Rudolf Usselmann ////
|
| 8 |
|
|
//// rudi@asics.ws ////
|
| 9 |
|
|
//// ////
|
| 10 |
|
|
//// Modified by Julius Baxter, July, 2010 ////
|
| 11 |
|
|
//// julius.baxter@orsoc.se ////
|
| 12 |
|
|
//// ////
|
| 13 |
|
|
//// TODO: Fix bug where 1.0f in round up goes to integer 2 ////
|
| 14 |
|
|
//// ////
|
| 15 |
|
|
/////////////////////////////////////////////////////////////////////
|
| 16 |
|
|
//// ////
|
| 17 |
|
|
//// Copyright (C) 2000 Rudolf Usselmann ////
|
| 18 |
|
|
//// rudi@asics.ws ////
|
| 19 |
|
|
//// ////
|
| 20 |
|
|
//// This source file may be used and distributed without ////
|
| 21 |
|
|
//// restriction provided that this copyright statement is not ////
|
| 22 |
|
|
//// removed from the file and that any derivative work contains ////
|
| 23 |
|
|
//// the original copyright notice and the associated disclaimer.////
|
| 24 |
|
|
//// ////
|
| 25 |
|
|
//// THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY ////
|
| 26 |
|
|
//// EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED ////
|
| 27 |
|
|
//// TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS ////
|
| 28 |
|
|
//// FOR A PARTICULAR PURPOSE. IN NO EVENT SHALL THE AUTHOR ////
|
| 29 |
|
|
//// OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, ////
|
| 30 |
|
|
//// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES ////
|
| 31 |
|
|
//// (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE ////
|
| 32 |
|
|
//// GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR ////
|
| 33 |
|
|
//// BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF ////
|
| 34 |
|
|
//// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT ////
|
| 35 |
|
|
//// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT ////
|
| 36 |
|
|
//// OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE ////
|
| 37 |
|
|
//// POSSIBILITY OF SUCH DAMAGE. ////
|
| 38 |
|
|
//// ////
|
| 39 |
|
|
/////////////////////////////////////////////////////////////////////
|
| 40 |
|
|
|
| 41 |
|
|
`timescale 1ns / 100ps
|
| 42 |
|
|
|
| 43 |
|
|
/*
|
| 44 |
|
|
|
| 45 |
|
|
FPU Operations (fpu_op):
|
| 46 |
|
|
========================
|
| 47 |
|
|
|
| 48 |
|
|
|
| 49 |
|
|
1 =
|
| 50 |
|
|
2 =
|
| 51 |
|
|
3 =
|
| 52 |
|
|
4 = int to float
|
| 53 |
|
|
5 = float to int
|
| 54 |
|
|
6 =
|
| 55 |
|
|
7 =
|
| 56 |
|
|
|
| 57 |
|
|
Rounding Modes (rmode):
|
| 58 |
|
|
=======================
|
| 59 |
|
|
|
| 60 |
|
|
|
| 61 |
|
|
1 = round_to_zero
|
| 62 |
|
|
2 = round_up
|
| 63 |
|
|
3 = round_down
|
| 64 |
|
|
|
| 65 |
|
|
*/
|
| 66 |
|
|
|
| 67 |
|
|
|
| 68 |
|
|
module or1200_fpu_intfloat_conv
|
| 69 |
|
|
(
|
| 70 |
|
|
clk, rmode, fpu_op, opa, out, snan, ine, inv,
|
| 71 |
|
|
overflow, underflow, zero
|
| 72 |
|
|
);
|
| 73 |
|
|
input clk;
|
| 74 |
|
|
input [1:0] rmode;
|
| 75 |
|
|
input [2:0] fpu_op;
|
| 76 |
|
|
input [31:0] opa;
|
| 77 |
|
|
output [31:0] out;
|
| 78 |
|
|
output snan;
|
| 79 |
|
|
output ine;
|
| 80 |
|
|
output inv;
|
| 81 |
|
|
output overflow;
|
| 82 |
|
|
output underflow;
|
| 83 |
|
|
output zero;
|
| 84 |
|
|
|
| 85 |
|
|
|
| 86 |
|
|
parameter INF = 31'h7f800000,
|
| 87 |
|
|
QNAN = 31'h7fc00001,
|
| 88 |
|
|
SNAN = 31'h7f800001;
|
| 89 |
|
|
|
| 90 |
|
|
////////////////////////////////////////////////////////////////////////
|
| 91 |
|
|
//
|
| 92 |
|
|
// Local Wires
|
| 93 |
|
|
//
|
| 94 |
|
|
reg zero;
|
| 95 |
|
|
reg [31:0] opa_r; // Input operand registers
|
| 96 |
|
|
reg [31:0] out; // Output register
|
| 97 |
|
|
reg div_by_zero; // Divide by zero output register
|
| 98 |
|
|
wire [7:0] exp_fasu; // Exponent output from EQU block
|
| 99 |
|
|
reg [7:0] exp_r; // Exponent output (registerd)
|
| 100 |
|
|
wire co; // carry output
|
| 101 |
|
|
wire [30:0] out_d; // Intermediate final result output
|
| 102 |
|
|
wire overflow_d, underflow_d;// Overflow/Underflow
|
| 103 |
|
|
reg inf, snan, qnan;// Output Registers for INF, S/QNAN
|
| 104 |
|
|
reg ine; // Output Registers for INE
|
| 105 |
|
|
reg [1:0] rmode_r1, rmode_r2,// Pipeline registers for round mode
|
| 106 |
|
|
rmode_r3;
|
| 107 |
|
|
reg [2:0] fpu_op_r1, fpu_op_r2,// Pipeline registers for fp
|
| 108 |
|
|
// operation
|
| 109 |
|
|
fpu_op_r3;
|
| 110 |
|
|
|
| 111 |
|
|
////////////////////////////////////////////////////////////////////////
|
| 112 |
|
|
//
|
| 113 |
|
|
// Input Registers
|
| 114 |
|
|
//
|
| 115 |
|
|
|
| 116 |
|
|
always @(posedge clk)
|
| 117 |
|
|
opa_r <= opa;
|
| 118 |
|
|
|
| 119 |
|
|
|
| 120 |
|
|
always @(posedge clk)
|
| 121 |
|
|
rmode_r1 <= rmode;
|
| 122 |
|
|
|
| 123 |
|
|
always @(posedge clk)
|
| 124 |
|
|
rmode_r2 <= rmode_r1;
|
| 125 |
|
|
|
| 126 |
|
|
always @(posedge clk)
|
| 127 |
|
|
rmode_r3 <= rmode_r2;
|
| 128 |
|
|
|
| 129 |
|
|
always @(posedge clk)
|
| 130 |
|
|
fpu_op_r1 <= fpu_op;
|
| 131 |
|
|
|
| 132 |
|
|
always @(posedge clk)
|
| 133 |
|
|
fpu_op_r2 <= fpu_op_r1;
|
| 134 |
|
|
|
| 135 |
|
|
always @(posedge clk)
|
| 136 |
|
|
fpu_op_r3 <= fpu_op_r2;
|
| 137 |
|
|
|
| 138 |
|
|
////////////////////////////////////////////////////////////////////////
|
| 139 |
|
|
//
|
| 140 |
|
|
// Exceptions block
|
| 141 |
|
|
//
|
| 142 |
|
|
wire inf_d, ind_d, qnan_d, snan_d, opa_nan;
|
| 143 |
|
|
wire opa_00;
|
| 144 |
|
|
wire opa_inf;
|
| 145 |
|
|
wire opa_dn;
|
| 146 |
|
|
|
| 147 |
|
|
or1200_fpu_intfloat_conv_except u0
|
| 148 |
|
|
( .clk(clk),
|
| 149 |
|
|
.opa(opa_r),
|
| 150 |
|
|
.opb(),
|
| 151 |
|
|
.inf(inf_d),
|
| 152 |
|
|
.ind(ind_d),
|
| 153 |
|
|
.qnan(qnan_d),
|
| 154 |
|
|
.snan(snan_d),
|
| 155 |
|
|
.opa_nan(opa_nan),
|
| 156 |
|
|
.opb_nan(),
|
| 157 |
|
|
.opa_00(opa_00),
|
| 158 |
|
|
.opb_00(),
|
| 159 |
|
|
.opa_inf(opa_inf),
|
| 160 |
|
|
.opb_inf(),
|
| 161 |
|
|
.opa_dn(opa_dn),
|
| 162 |
|
|
.opb_dn()
|
| 163 |
|
|
);
|
| 164 |
|
|
|
| 165 |
|
|
////////////////////////////////////////////////////////////////////////
|
| 166 |
|
|
//
|
| 167 |
|
|
// Pre-Normalize block
|
| 168 |
|
|
// - Adjusts the numbers to equal exponents and sorts them
|
| 169 |
|
|
// - determine result sign
|
| 170 |
|
|
// - determine actual operation to perform (add or sub)
|
| 171 |
|
|
//
|
| 172 |
|
|
|
| 173 |
|
|
wire nan_sign_d, result_zero_sign_d;
|
| 174 |
|
|
reg sign_fasu_r;
|
| 175 |
|
|
wire [1:0] exp_ovf;
|
| 176 |
|
|
reg [1:0] exp_ovf_r;
|
| 177 |
|
|
|
| 178 |
|
|
// This is all we need from post-norm module for int-float conversion
|
| 179 |
|
|
reg opa_sign_r;
|
| 180 |
|
|
always @(posedge clk)
|
| 181 |
|
|
opa_sign_r <= opa_r[31];
|
| 182 |
|
|
|
| 183 |
|
|
always @(posedge clk)
|
| 184 |
|
|
sign_fasu_r <= opa_sign_r; //sign_fasu;
|
| 185 |
|
|
|
| 186 |
|
|
|
| 187 |
|
|
////////////////////////////////////////////////////////////////////////
|
| 188 |
|
|
//
|
| 189 |
|
|
// Normalize Result
|
| 190 |
|
|
//
|
| 191 |
|
|
wire ine_d;
|
| 192 |
|
|
wire inv_d;
|
| 193 |
|
|
wire sign_d;
|
| 194 |
|
|
reg sign;
|
| 195 |
|
|
reg [30:0] opa_r1;
|
| 196 |
|
|
reg [47:0] fract_i2f;
|
| 197 |
|
|
reg opas_r1, opas_r2;
|
| 198 |
|
|
wire f2i_out_sign;
|
| 199 |
|
|
wire [47:0] fract_denorm;
|
| 200 |
|
|
|
| 201 |
|
|
always @(posedge clk) // Exponent must be once cycle delayed
|
| 202 |
|
|
case(fpu_op_r2)
|
| 203 |
|
|
//4: exp_r <= 0;
|
| 204 |
|
|
5: exp_r <= opa_r1[30:23];
|
| 205 |
|
|
default: exp_r <= 0;
|
| 206 |
|
|
endcase
|
| 207 |
|
|
|
| 208 |
|
|
always @(posedge clk)
|
| 209 |
|
|
opa_r1 <= opa_r[30:0];
|
| 210 |
|
|
|
| 211 |
|
|
always @(posedge clk)
|
| 212 |
|
|
fract_i2f <= (fpu_op_r2==5) ?
|
| 213 |
|
|
(sign_d ? 1-{24'h00, (|opa_r1[30:23]), opa_r1[22:0]}-1 :
|
| 214 |
|
|
{24'h0, (|opa_r1[30:23]), opa_r1[22:0]})
|
| 215 |
|
|
: (sign_d ? 1 - {opa_r1, 17'h01} : {opa_r1, 17'h0});
|
| 216 |
|
|
|
| 217 |
|
|
assign fract_denorm = fract_i2f;
|
| 218 |
|
|
|
| 219 |
|
|
always @(posedge clk)
|
| 220 |
|
|
opas_r1 <= opa_r[31];
|
| 221 |
|
|
|
| 222 |
|
|
always @(posedge clk)
|
| 223 |
|
|
opas_r2 <= opas_r1;
|
| 224 |
|
|
|
| 225 |
|
|
assign sign_d = opa_sign_r; //sign_fasu;
|
| 226 |
|
|
|
| 227 |
|
|
always @(posedge clk)
|
| 228 |
|
|
sign <= (rmode_r2==2'h3) ? !sign_d : sign_d;
|
| 229 |
|
|
|
| 230 |
|
|
|
| 231 |
|
|
// Special case of largest negative integer we can convert to - usually
|
| 232 |
|
|
// gets picked up as invalid, but really it's not, so deal with it as a
|
| 233 |
|
|
// special case.
|
| 234 |
|
|
wire f2i_special_case_no_inv;
|
| 235 |
|
|
assign f2i_special_case_no_inv = (opa == 32'hcf000000);
|
| 236 |
|
|
|
| 237 |
|
|
|
| 238 |
|
|
or1200_fpu_post_norm_intfloat_conv u4
|
| 239 |
|
|
(
|
| 240 |
|
|
.clk(clk), // System Clock
|
| 241 |
|
|
.fpu_op(fpu_op_r3), // Floating Point Operation
|
| 242 |
|
|
.opas(opas_r2), // OPA Sign
|
| 243 |
|
|
.sign(sign), // Sign of the result
|
| 244 |
|
|
.rmode(rmode_r3), // Rounding mode
|
| 245 |
|
|
.fract_in(fract_denorm), // Fraction Input
|
| 246 |
|
|
|
| 247 |
|
|
.exp_in(exp_r), // Exponent Input
|
| 248 |
|
|
.opa_dn(opa_dn), // Operand A Denormalized
|
| 249 |
|
|
.opa_nan(opa_nan),
|
| 250 |
|
|
.opa_inf(opa_inf),
|
| 251 |
|
|
|
| 252 |
|
|
.opb_dn(), // Operand B Denormalized
|
| 253 |
|
|
.out(out_d), // Normalized output (un-registered)
|
| 254 |
|
|
.ine(ine_d), // Result Inexact output (un-registered)
|
| 255 |
|
|
.inv(inv_d), // Invalid input for f2i operation
|
| 256 |
|
|
.overflow(overflow_d), // Overflow output (un-registered)
|
| 257 |
|
|
.underflow(underflow_d),// Underflow output (un-registered)
|
| 258 |
|
|
.f2i_out_sign(f2i_out_sign) // F2I Output Sign
|
| 259 |
|
|
);
|
| 260 |
|
|
|
| 261 |
|
|
////////////////////////////////////////////////////////////////////////
|
| 262 |
|
|
//
|
| 263 |
|
|
// FPU Outputs
|
| 264 |
|
|
//
|
| 265 |
|
|
reg fasu_op_r1, fasu_op_r2;
|
| 266 |
|
|
wire [30:0] out_fixed;
|
| 267 |
|
|
wire output_zero_fasu;
|
| 268 |
|
|
wire overflow_fasu;
|
| 269 |
|
|
wire out_d_00;
|
| 270 |
|
|
wire ine_fasu;
|
| 271 |
|
|
wire underflow_fasu;
|
| 272 |
|
|
|
| 273 |
|
|
|
| 274 |
|
|
/*
|
| 275 |
|
|
always @(posedge clk)
|
| 276 |
|
|
fasu_op_r1 <= fasu_op;
|
| 277 |
|
|
|
| 278 |
|
|
always @(posedge clk)
|
| 279 |
|
|
fasu_op_r2 <= fasu_op_r1;
|
| 280 |
|
|
*/
|
| 281 |
|
|
// Force pre-set values for non numerical output
|
| 282 |
|
|
|
| 283 |
|
|
assign out_fixed = ( (qnan_d | snan_d) |
|
| 284 |
|
|
(ind_d /*& !fasu_op_r2*/)) ? QNAN : INF;
|
| 285 |
|
|
|
| 286 |
|
|
always @(posedge clk)
|
| 287 |
|
|
out[30:0] <= /*((inf_d & (fpu_op_r3!=3'b101)) | snan_d | qnan_d)
|
| 288 |
|
|
& fpu_op_r3!=3'b100 ? out_fixed :*/ out_d;
|
| 289 |
|
|
|
| 290 |
|
|
assign out_d_00 = !(|out_d);
|
| 291 |
|
|
|
| 292 |
|
|
|
| 293 |
|
|
always @(posedge clk)
|
| 294 |
|
|
out[31] <= (fpu_op_r3==3'b101) ?
|
| 295 |
|
|
f2i_out_sign : sign_fasu_r;
|
| 296 |
|
|
|
| 297 |
|
|
|
| 298 |
|
|
|
| 299 |
|
|
// Exception Outputs
|
| 300 |
|
|
assign ine_fasu = (ine_d | overflow_d | underflow_d) &
|
| 301 |
|
|
!(snan_d | qnan_d | inf_d);
|
| 302 |
|
|
|
| 303 |
|
|
always @(posedge clk)
|
| 304 |
|
|
ine <= fpu_op_r3[2] ? ine_d : ine_fasu;
|
| 305 |
|
|
|
| 306 |
|
|
assign overflow = overflow_d & !(snan_d | qnan_d | inf_d);
|
| 307 |
|
|
assign underflow = underflow_d & !(inf_d | snan_d | qnan_d);
|
| 308 |
|
|
|
| 309 |
|
|
always @(posedge clk)
|
| 310 |
|
|
snan <= snan_d & (fpu_op_r3==3'b101); // Only signal sNaN when ftoi
|
| 311 |
|
|
|
| 312 |
|
|
// Status Outputs
|
| 313 |
|
|
assign output_zero_fasu = out_d_00 & !(inf_d | snan_d | qnan_d);
|
| 314 |
|
|
|
| 315 |
|
|
always @(posedge clk)
|
| 316 |
|
|
zero <= fpu_op_r3==3'b101 ? out_d_00 & !(snan_d | qnan_d) :
|
| 317 |
|
|
output_zero_fasu ;
|
| 318 |
|
|
assign inv = inv_d & !f2i_special_case_no_inv;
|
| 319 |
|
|
|
| 320 |
|
|
endmodule // or1200_fpu_intfloat_conv
|